cpufreq: Skip cpufreq resume if it's not suspended
authorBo Yan <byan@nvidia.com>
Tue, 23 Jan 2018 21:57:55 +0000 (13:57 -0800)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Mon, 5 Feb 2018 10:03:33 +0000 (11:03 +0100)
commit703cbaa601ff3fb554d1246c336ba727cc083ea0
tree1d6c54f84f12ed6de8b299cb6b5976c504e40550
parent7f3fdd40a7dfaa7405185250974b0fabd08c1f8b
cpufreq: Skip cpufreq resume if it's not suspended

cpufreq_resume can be called even without preceding cpufreq_suspend.
This can happen in following scenario:

    suspend_devices_and_enter
       --> dpm_suspend_start
          --> dpm_prepare
              --> device_prepare : this function errors out
          --> dpm_suspend: this is skipped due to dpm_prepare failure
                           this means cpufreq_suspend is skipped over
       --> goto Recover_platform, due to previous error
       --> goto Resume_devices
       --> dpm_resume_end
           --> dpm_resume
               --> cpufreq_resume

In case schedutil is used as frequency governor, cpufreq_resume will
eventually call sugov_start, which does following:

    memset(sg_cpu, 0, sizeof(*sg_cpu));
    ....

This effectively erases function pointer for frequency update, causing
crash later on. The function pointer would have been set correctly if
subsequent cpufreq_add_update_util_hook runs successfully, but that
function returns earlier because cpufreq_suspend was not called:

    if (WARN_ON(per_cpu(cpufreq_update_util_data, cpu)))
return;

The fix is to check cpufreq_suspended first, if it's false, that means
cpufreq_suspend was not called in the first place, so do not resume
cpufreq.

Signed-off-by: Bo Yan <byan@nvidia.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Dropped printing a message ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/cpufreq/cpufreq.c